|
/* Macros which must be defined outside:
NAME : name for this instance of the code
WORD : type for a word, please make it unsigned!
PTRINT : numerical type which can hold a pointer
WPR : words per row
MATROWS : rows in one matrix, must be equal to WPR*sizeof(WORD)*8
ALIGN : alignment of the arena, set e.g. to 0x100000L to align on 1MB
GREASE : grease level, must be a divisor of sizeof(WORD)*8
GREASE2 : must be 2^GREASE
*/
#include <stdlib.h>
#include <string.h>
#define WORDLEN sizeof(WORD)
#define WORDBITS (sizeof(WORD)*8)
#define MATMEM (MATROWS*WPR*sizeof(WORD))
#define GREASEMASK (GREASE2-1)
WORD *ARENA;
WORD *REGS[128]; /* Pointers to the matrix registers */
WORD *GRACCU;
/* Number of standard matrix registers: */
int NRREGS; /* This is set by gf2_allocmem/gf2_usemem */
int WPRNAME = WPR;
int gf2_usemem(void *memspace, PTRINT cachesize)
{
int i;
PTRINT p;
ARENA = memspace;
NRREGS = (int) (((4*cachesize)/5) / MATMEM - 32);
if (NRREGS < 8) return -1;
if (NRREGS > 128) NRREGS = 128;
p = (PTRINT) ARENA;
for (i = 0;i < NRREGS;i++) {
REGS[i] = (WORD *) p;
p += MATMEM;
}
GRACCU = (WORD *) p;
return 0;
}
int gf2_allocmem(PTRINT cachesize)
{
ARENA = (WORD *) malloc(cachesize+ALIGN);
if (ARENA == NULL) return -2;
ARENA = (WORD *) (((PTRINT) ARENA + ALIGN - 1) & (~(ALIGN-1)));
return gf2_usemem(ARENA,cachesize);
}
#if WPR == 1
#define ROWCLEAR(r) *r++ = 0
#else
#define ROWCLEAR(r) do { int i; for (i = WPR;i > 0;i--) *r++ = 0; } while(0)
#endif
#if WPR == 1
#define ROWADD3(r,p,q) *r++ = *p++ ^ *q++
#else
#define ROWADD3(r,p,q) do \
{ int i; for (i = WPR;i > 0;i--) *r++ = *p++ ^ *q++;} while(0)
#endif
#if WPR == 1
#define ROWADD2(r,q) *r++ ^= *q++
#else
#define ROWADD2(r,q) do \
{ int i; for (i = WPR;i > 0;i--) { *r = *r ^ *q++; r++; } } while(0)
#endif
#if GREASE == 4 && WPR == 1
/* In this case words is always equal to 1! */
void gf2_grease(int a, int words)
/* GREASE a, which is a register number
words must be <= WPR and gf2_mul must then
be called with the same value! */
{
WORD *r,*w;
int k;
r = REGS[a];
w = GRACCU;
for (k = (sizeof(WORD)*8/GREASE);k > 0;k--) {
*w++ = 0;
*w++ = r[0];
*w++ = r[1];
*w++ = r[0] ^ r[1];
*w++ = r[2];
*w++ = r[2] ^ r[0];
*w++ = r[2] ^ r[1];
*w++ = r[2] ^ r[1] ^ r[0];
*w++ = r[3];
*w = r[3] ^ w[-8]; w++;
*w = r[3] ^ w[-8]; w++;
*w = r[3] ^ w[-8]; w++;
*w = r[3] ^ w[-8]; w++;
*w = r[3] ^ w[-8]; w++;
*w = r[3] ^ w[-8]; w++;
*w = r[3] ^ w[-8]; w++;
r += 4;
}
}
#else
void gf2_grease(int a, int words)
/* GREASE a, which is a register number
words must be <= WPR and gf2_mul must then
be called with the same value! */
{
WORD *r,*q;
int j;
int l;
int i;
int k;
WORD *p,*rr;
p = REGS[a];
r = GRACCU;
for (k = words*(sizeof(WORD)*8/GREASE);k > 0;k--) {
rr = r;
ROWCLEAR(r);
l = 1;
for (i = GREASE;i > 0;i--) {
q = rr;
for (j = l;j > 0;j--) {
ROWADD3(r,q,p);
p -= WPR;
}
p += WPR;
l <<= 1;
}
}
}
#endif
void gf2_mul(int d, int a, int rows, int words)
/* d := a * greaseaccu
d and a are register numbers, d must be different from a!
rows is the number of rows taken in left matrix a
words*sizeof(WORD)*8 is the number of rows taken in right matrix
(--> greaseaccu)
Note that words must be <= WPR and gf2_grease must have been called
with the same value! */
{
WORD *r,*p,*q;
int i,j,k;
WORD w;
r = REGS[d];
p = REGS[a];
if (words == 1 ) {
for (i = rows;i > 0;i--) {
q = GRACCU;
ROWCLEAR(r);
r -= WPR;
w = *p++;
if (w) {
#if WPR == 1
WORD w2 = 0;
for (j = sizeof(WORD)*8/GREASE;j > 0;j--) {
if (q[w & GREASEMASK]) w2 ^= q[w & GREASEMASK];
w >>= GREASE;
q += GREASE2*WPR;
}
*r = w2;
#else
for (j = sizeof(WORD)*8/GREASE;j > 0;j--) {
WORD *qq = q + WPR * (w & GREASEMASK);
ROWADD2(r,qq);
r -= WPR;
w >>= GREASE;
q += GREASE2*WPR;
}
#endif
} else q += GREASE2*WPR*(sizeof(WORD)*8/GREASE);
p += WPR-words;
r += WPR;
}
} else {
for (i = rows;i > 0;i--) {
q = GRACCU;
ROWCLEAR(r);
r -= WPR;
for (k = words;k > 0;k--) {
w = *p++;
if (w) {
#if WPR == 1
WORD w2 = 0;
for (j = sizeof(WORD)*8/GREASE;j > 0;j--) {
if (q[w & GREASEMASK]) w2 ^= q[w & GREASEMASK];
w >>= GREASE;
q += GREASE2*WPR;
}
*r = w2;
#else
for (j = sizeof(WORD)*8/GREASE;j > 0;j--) {
WORD *qq = q + WPR * (w & GREASEMASK);
ROWADD2(r,qq);
r -= WPR;
w >>= GREASE;
q += GREASE2*WPR;
}
#endif
} else q += GREASE2*WPR*(sizeof(WORD)*8/GREASE);
}
p += WPR-words;
r += WPR;
}
}
}
void gf2_zero(int d)
{
int i;
WORD *p = REGS[d];
for (i = MATROWS;i > 0;i--) ROWCLEAR(p);
}
void gf2_add(int d, int a, int b)
/* d := a + b all are register numbers */
{
WORD *p, *q, *r;
int i;
p = REGS[a]; q = REGS[b]; r = REGS[d];
for (i = MATROWS;i > 0;i--) ROWADD3(r,p,q);
}
void gf2_copy(int d, int a)
/* d := a all are register numbers */
{
memcpy(REGS[d],REGS[a],MATMEM);
}
#undef ROWCLEAR
#undef ROWADD2
#undef ROWADD3
#undef ARENA
#undef _ARENA
#undef __ARENA
#undef REGS
#undef _REGS
#undef __REGS
#undef GRACCU
#undef _GRACCU
#undef __GRACCU
#undef NRREGS
#undef _NRREGS
#undef __NRREGS
#undef gf2_allocmem
#undef _gf2_allocmem
#undef __gf2_allocmem
#undef gf2_usemem
#undef _gf2_usemem
#undef __gf2_usemem
#undef gf2_grease
#undef _gf2_grease
#undef __gf2_grease
#undef gf2_mul
#undef _gf2_mul
#undef __gf2_mul
#undef gf2_zero
#undef _gf2_zero
#undef __gf2_zero
#undef gf2_add
#undef _gf2_add
#undef __gf2_add
#undef gf2_copy
#undef _gf2_copy
#undef __gf2_copy
#undef WORDLEN
#undef WORDBITS
#undef ARENA
#undef REGS
#undef GRACCU
#undef MATMEM
[ Dauer der Verarbeitung: 0.17 Sekunden
(vorverarbeitet)
]
|
